diff --git a/Android.mk b/Android.mk index 6e5f3263..3aa44c4b 100644 --- a/Android.mk +++ b/Android.mk @@ -44,6 +44,7 @@ LOCAL_SRC_FILES := \ src/dsp/enc_mips32.c \ src/dsp/enc_neon.$(NEON) \ src/dsp/enc_sse2.c \ + src/dsp/filters_mips_dsp_r2.c \ src/dsp/lossless.c \ src/dsp/lossless_mips32.c \ src/dsp/lossless_neon.$(NEON) \ diff --git a/Makefile.vc b/Makefile.vc index 4de65104..8dbf275c 100644 --- a/Makefile.vc +++ b/Makefile.vc @@ -178,6 +178,7 @@ DSP_DEC_OBJS = \ $(DIROBJ)\dsp\dec_mips32.obj \ $(DIROBJ)\dsp\dec_neon.obj \ $(DIROBJ)\dsp\dec_sse2.obj \ + $(DIROBJ)\dsp\filters_mips_dsp_r2.obj \ $(DIROBJ)\dsp\lossless.obj \ $(DIROBJ)\dsp\lossless_mips32.obj \ $(DIROBJ)\dsp\lossless_neon.obj \ diff --git a/makefile.unix b/makefile.unix index 77ee3d0d..1e4f2d57 100644 --- a/makefile.unix +++ b/makefile.unix @@ -114,6 +114,7 @@ DSP_DEC_OBJS = \ src/dsp/dec_mips32.o \ src/dsp/dec_neon.o \ src/dsp/dec_sse2.o \ + src/dsp/filters_mips_dsp_r2.o \ src/dsp/lossless.o \ src/dsp/lossless_mips32.o \ src/dsp/lossless_neon.o \ diff --git a/src/dsp/Makefile.am b/src/dsp/Makefile.am index 485c57f3..e1c6e887 100644 --- a/src/dsp/Makefile.am +++ b/src/dsp/Makefile.am @@ -16,6 +16,7 @@ COMMON_SOURCES += dec_clip_tables.c COMMON_SOURCES += dec_mips32.c COMMON_SOURCES += dec_neon.c COMMON_SOURCES += dsp.h +COMMON_SOURCES += filters_mips_dsp_r2.c COMMON_SOURCES += lossless.c COMMON_SOURCES += lossless.h COMMON_SOURCES += lossless_mips32.c diff --git a/src/dsp/alpha_processing.c b/src/dsp/alpha_processing.c index 09deacfb..3e1b634b 100644 --- a/src/dsp/alpha_processing.c +++ b/src/dsp/alpha_processing.c @@ -290,9 +290,20 @@ void (*WebPApplyAlphaMultiply4444)(uint8_t*, int, int, int); //------------------------------------------------------------------------------ // Init function +extern void VP8FiltersInitMIPSdspR2(void); + void WebPInitAlphaProcessing(void) { WebPMultARGBRow = MultARGBRow; WebPMultRow = MultRow; WebPApplyAlphaMultiply = ApplyAlphaMultiply; WebPApplyAlphaMultiply4444 = ApplyAlphaMultiply_16b; + + // If defined, use CPUInfo() to overwrite some pointers with faster versions. + if (VP8GetCPUInfo != NULL) { +#if defined(WEBP_USE_MIPS_DSP_R2) + if (VP8GetCPUInfo(kMIPSdspR2)) { + VP8FiltersInitMIPSdspR2(); + } +#endif + } } diff --git a/src/dsp/filters_mips_dsp_r2.c b/src/dsp/filters_mips_dsp_r2.c new file mode 100644 index 00000000..d9e554ff --- /dev/null +++ b/src/dsp/filters_mips_dsp_r2.c @@ -0,0 +1,331 @@ +// Copyright 2014 Google Inc. All Rights Reserved. +// +// Use of this source code is governed by a BSD-style license +// that can be found in the COPYING file in the root of the source +// tree. An additional intellectual property rights grant can be found +// in the file PATENTS. All contributing project authors may +// be found in the AUTHORS file in the root of the source tree. +// ----------------------------------------------------------------------------- +// +// Spatial prediction using various filters +// +// Author(s): Branimir Vasic (branimir.vasic@imgtec.com) +// Djordje Pesut (djordje.pesut@imgtec.com) + +#include "./dsp.h" + +#if defined(WEBP_USE_MIPS_DSP_R2) + +#include "../utils/filters.h" +#include +#include +#include + +//------------------------------------------------------------------------------ +// Helpful macro. + +# define SANITY_CHECK(in, out) \ + assert(in != NULL); \ + assert(out != NULL); \ + assert(width > 0); \ + assert(height > 0); \ + assert(stride >= width); \ + assert(row >= 0 && num_rows > 0 && row + num_rows <= height); \ + (void)height; // Silence unused warning. + +#define DO_PREDICT_LINE(SRC, PRED, DST, LENGTH, INVERSE) do { \ + const uint8_t* psrc = (uint8_t*)(SRC); \ + const uint8_t* ppred = (uint8_t*)(PRED); \ + uint8_t* pdst = (uint8_t*)(DST); \ + const int ilength = (int)(LENGTH); \ + int temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7; \ + __asm__ volatile ( \ + ".set push \n\t" \ + ".set noreorder \n\t" \ + "srl %[temp0], %[length], 0x3 \n\t" \ + "beqz %[temp0], 2f \n\t" \ + " andi %[temp7], %[length], 0x7 \n\t" \ + "1: \n\t" \ + "ulw %[temp1], 0(%[src]) \n\t" \ + "ulw %[temp2], 0(%[pred]) \n\t" \ + "ulw %[temp3], 4(%[src]) \n\t" \ + "ulw %[temp4], 4(%[pred]) \n\t" \ + "addiu %[src], %[src], 8 \n\t" \ + ".if "#INVERSE" \n\t" \ + "addu.qb %[temp5], %[temp1], %[temp2] \n\t" \ + "addu.qb %[temp6], %[temp3], %[temp4] \n\t" \ + ".else \n\t" \ + "subu.qb %[temp5], %[temp1], %[temp2] \n\t" \ + "subu.qb %[temp6], %[temp3], %[temp4] \n\t" \ + ".endif \n\t" \ + "addiu %[pred], %[pred], 8 \n\t" \ + "usw %[temp5], 0(%[dst]) \n\t" \ + "usw %[temp6], 4(%[dst]) \n\t" \ + "addiu %[temp0], %[temp0], -1 \n\t" \ + "bnez %[temp0], 1b \n\t" \ + " addiu %[dst], %[dst], 8 \n\t" \ + "beqz %[temp7], 3f \n\t" \ + " nop \n\t" \ + "2: \n\t" \ + "lbu %[temp1], 0(%[src]) \n\t" \ + "lbu %[temp2], 0(%[pred]) \n\t" \ + "addiu %[src], %[src], 1 \n\t" \ + "addiu %[pred], %[pred], 1 \n\t" \ + ".if "#INVERSE" \n\t" \ + "addu %[temp3], %[temp1], %[temp2] \n\t" \ + ".else \n\t" \ + "subu %[temp3], %[temp1], %[temp2] \n\t" \ + ".endif \n\t" \ + "sb %[temp3], 0(%[dst]) \n\t" \ + "addiu %[temp7], %[temp7], -1 \n\t" \ + "bnez %[temp7], 2b \n\t" \ + " addiu %[dst], %[dst], 1 \n\t" \ + "3: \n\t" \ + ".set pop \n\t" \ + : [temp0]"=&r"(temp0), [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), \ + [temp3]"=&r"(temp3), [temp4]"=&r"(temp4), [temp5]"=&r"(temp5), \ + [temp6]"=&r"(temp6), [temp7]"=&r"(temp7), [pred]"+r"(ppred), \ + [dst]"+r"(pdst), [src]"+r"(psrc) \ + : [length]"r"(ilength) \ + : "memory" \ + ); \ + } while (0) + +static WEBP_INLINE void PredictLine(const uint8_t* src, const uint8_t* pred, + uint8_t* dst, int length, int inverse) { + if (inverse) { + DO_PREDICT_LINE(src, pred, dst, length, 1); + } else { + DO_PREDICT_LINE(src, pred, dst, length, 0); + } +} + +#define PREDICT_LINE_ONE_PASS(SRC, PRED, DST, INVERSE) do { \ + int temp1, temp2, temp3; \ + __asm__ volatile ( \ + "lbu %[temp1], 0(%[src]) \n\t" \ + "lbu %[temp2], 0(%[pred]) \n\t" \ + ".if "#INVERSE" \n\t" \ + "addu %[temp3], %[temp1], %[temp2] \n\t" \ + ".else \n\t" \ + "subu %[temp3], %[temp1], %[temp2] \n\t" \ + ".endif \n\t" \ + "sb %[temp3], 0(%[dst]) \n\t" \ + : [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), [temp3]"=&r"(temp3) \ + : [pred]"r"((PRED)), [dst]"r"((DST)), [src]"r"((SRC)) \ + : "memory" \ + ); \ + } while (0) + +//------------------------------------------------------------------------------ +// Horizontal filter. + +#define FILTER_LINE_BY_LINE(INVERSE) do { \ + while (row < last_row) { \ + PREDICT_LINE_ONE_PASS(in, preds - stride, out, INVERSE); \ + DO_PREDICT_LINE(in + 1, preds, out + 1, width - 1, INVERSE); \ + ++row; \ + preds += stride; \ + in += stride; \ + out += stride; \ + } \ + } while (0) + +static WEBP_INLINE void DoHorizontalFilter(const uint8_t* in, + int width, int height, int stride, + int row, int num_rows, + int inverse, uint8_t* out) { + const uint8_t* preds; + const size_t start_offset = row * stride; + const int last_row = row + num_rows; + SANITY_CHECK(in, out); + in += start_offset; + out += start_offset; + preds = inverse ? out : in; + + if (row == 0) { + // Leftmost pixel is the same as input for topmost scanline. + out[0] = in[0]; + PredictLine(in + 1, preds, out + 1, width - 1, inverse); + row = 1; + preds += stride; + in += stride; + out += stride; + } + + // Filter line-by-line. + if (inverse) { + FILTER_LINE_BY_LINE(1); + } else { + FILTER_LINE_BY_LINE(0); + } +} + +#undef FILTER_LINE_BY_LINE + +static void HorizontalFilter(const uint8_t* data, int width, int height, + int stride, uint8_t* filtered_data) { + DoHorizontalFilter(data, width, height, stride, 0, height, 0, filtered_data); +} + +static void HorizontalUnfilter(int width, int height, int stride, int row, + int num_rows, uint8_t* data) { + DoHorizontalFilter(data, width, height, stride, row, num_rows, 1, data); +} + +//------------------------------------------------------------------------------ +// Vertical filter. + +#define FILTER_LINE_BY_LINE(INVERSE) do { \ + while (row < last_row) { \ + DO_PREDICT_LINE(in, preds, out, width, INVERSE); \ + ++row; \ + preds += stride; \ + in += stride; \ + out += stride; \ + } \ + } while (0) + +static WEBP_INLINE void DoVerticalFilter(const uint8_t* in, + int width, int height, int stride, + int row, int num_rows, + int inverse, uint8_t* out) { + const uint8_t* preds; + const size_t start_offset = row * stride; + const int last_row = row + num_rows; + SANITY_CHECK(in, out); + in += start_offset; + out += start_offset; + preds = inverse ? out : in; + + if (row == 0) { + // Very first top-left pixel is copied. + out[0] = in[0]; + // Rest of top scan-line is left-predicted. + PredictLine(in + 1, preds, out + 1, width - 1, inverse); + row = 1; + in += stride; + out += stride; + } else { + // We are starting from in-between. Make sure 'preds' points to prev row. + preds -= stride; + } + + // Filter line-by-line. + if (inverse) { + FILTER_LINE_BY_LINE(1); + } else { + FILTER_LINE_BY_LINE(0); + } +} + +#undef FILTER_LINE_BY_LINE + +static void VerticalFilter(const uint8_t* data, int width, int height, + int stride, uint8_t* filtered_data) { + DoVerticalFilter(data, width, height, stride, 0, height, 0, filtered_data); +} + +static void VerticalUnfilter(int width, int height, int stride, int row, + int num_rows, uint8_t* data) { + DoVerticalFilter(data, width, height, stride, row, num_rows, 1, data); +} + +//------------------------------------------------------------------------------ +// Gradient filter. + +static WEBP_INLINE int GradientPredictor(uint8_t a, uint8_t b, uint8_t c) { + int temp0; + __asm__ volatile ( + "addu %[temp0], %[a], %[b] \n\t" + "subu %[temp0], %[temp0], %[c] \n\t" + "shll_s.w %[temp0], %[temp0], 23 \n\t" + "precrqu_s.qb.ph %[temp0], %[temp0], $zero \n\t" + "srl %[temp0], %[temp0], 24 \n\t" + : [temp0]"=&r"(temp0) + : [a]"r"(a),[b]"r"(b),[c]"r"(c) + ); + return temp0; +} + +#define FILTER_LINE_BY_LINE(INVERSE, PREDS, OPERATION) do { \ + while (row < last_row) { \ + int w; \ + PREDICT_LINE_ONE_PASS(in, PREDS - stride, out, INVERSE); \ + for (w = 1; w < width; ++w) { \ + const int pred = GradientPredictor(PREDS[w - 1], \ + PREDS[w - stride], \ + PREDS[w - stride - 1]); \ + out[w] = in[w] OPERATION pred; \ + } \ + ++row; \ + in += stride; \ + out += stride; \ + } \ + } while (0) + +static WEBP_INLINE void DoGradientFilter(const uint8_t* in, + int width, int height, int stride, + int row, int num_rows, + int inverse, uint8_t* out) { + const uint8_t* preds; + const size_t start_offset = row * stride; + const int last_row = row + num_rows; + SANITY_CHECK(in, out); + in += start_offset; + out += start_offset; + preds = inverse ? out : in; + + // left prediction for top scan-line + if (row == 0) { + out[0] = in[0]; + PredictLine(in + 1, preds, out + 1, width - 1, inverse); + row = 1; + preds += stride; + in += stride; + out += stride; + } + + // Filter line-by-line. + if (inverse) { + FILTER_LINE_BY_LINE(1, out, +); + } else { + FILTER_LINE_BY_LINE(0, in, -); + } +} + +#undef FILTER_LINE_BY_LINE + +static void GradientFilter(const uint8_t* data, int width, int height, + int stride, uint8_t* filtered_data) { + DoGradientFilter(data, width, height, stride, 0, height, 0, filtered_data); +} + +static void GradientUnfilter(int width, int height, int stride, int row, + int num_rows, uint8_t* data) { + DoGradientFilter(data, width, height, stride, row, num_rows, 1, data); +} + +#undef PREDICT_LINE_ONE_PASS +#undef DO_PREDICT_LINE +#undef SANITY_CHECK + +#endif // WEBP_USE_MIPS_DSP_R2 + +//------------------------------------------------------------------------------ + +extern void VP8FiltersInitMIPSdspR2(void); + +void VP8FiltersInitMIPSdspR2(void) { +#if defined(WEBP_USE_MIPS_DSP_R2) + WebPFilters[WEBP_FILTER_HORIZONTAL] = HorizontalFilter; + WebPFilters[WEBP_FILTER_VERTICAL] = VerticalFilter; + WebPFilters[WEBP_FILTER_GRADIENT] = GradientFilter; + + WebPUnfilters[WEBP_FILTER_HORIZONTAL] = HorizontalUnfilter; + WebPUnfilters[WEBP_FILTER_VERTICAL] = VerticalUnfilter; + WebPUnfilters[WEBP_FILTER_GRADIENT] = GradientUnfilter; +#endif // WEBP_USE_MIPS_DSP_R2 +} + +//------------------------------------------------------------------------------ diff --git a/src/utils/filters.c b/src/utils/filters.c index 2d15bd0e..d9dae8e0 100644 --- a/src/utils/filters.c +++ b/src/utils/filters.c @@ -248,14 +248,14 @@ WEBP_FILTER_TYPE EstimateBestFilter(const uint8_t* data, //------------------------------------------------------------------------------ -const WebPFilterFunc WebPFilters[WEBP_FILTER_LAST] = { +WebPFilterFunc WebPFilters[WEBP_FILTER_LAST] = { NULL, // WEBP_FILTER_NONE HorizontalFilter, // WEBP_FILTER_HORIZONTAL VerticalFilter, // WEBP_FILTER_VERTICAL GradientFilter // WEBP_FILTER_GRADIENT }; -const WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST] = { +WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST] = { NULL, // WEBP_FILTER_NONE HorizontalUnfilter, // WEBP_FILTER_HORIZONTAL VerticalUnfilter, // WEBP_FILTER_VERTICAL diff --git a/src/utils/filters.h b/src/utils/filters.h index dde39cb5..fd680eae 100644 --- a/src/utils/filters.h +++ b/src/utils/filters.h @@ -41,12 +41,12 @@ typedef void (*WebPUnfilterFunc)(int width, int height, int stride, // in raster order. // 'stride' is number of bytes per scan line (with possible padding). // 'out' should be pre-allocated. -extern const WebPFilterFunc WebPFilters[WEBP_FILTER_LAST]; +extern WebPFilterFunc WebPFilters[WEBP_FILTER_LAST]; // In-place reconstruct the original data from the given filtered data. // The reconstruction will be done for 'num_rows' rows starting from 'row' // (assuming rows upto 'row - 1' are already reconstructed). -extern const WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST]; +extern WebPUnfilterFunc WebPUnfilters[WEBP_FILTER_LAST]; // Fast estimate of a potentially good filter. WEBP_FILTER_TYPE EstimateBestFilter(const uint8_t* data,